使用Spark计算PV、UV

标签：spark

日志字段格式：

id,ip,url,ref,cookie,time_stamp

把日志文件放到HDFS。仅取了1000行。

hadoop fs -put 1000_log hdfs://localhost:9000/user/root/input

直接在Scala Shell中读取文件并计算PV。

scala> val textFile = sc.textFile("hdfs://localhost:9000/user/root/input/1000_log")
scala> val textRDD = textFile.map(_.split("\t")).filter(_.length == 6)
scala> val result = textRDD.map(w => ((new java.net.URL(w(2))).getHost,1)).reduceByKey(_ + _).map(item => item.swap).sortByKey(false).map(item => item.swap)
scala> result.saveAsTextFile("hdfs://localhost:9000/user/root/out8.txt")

hadoop fs -get hdfs://localhost:9000/user/root/out8.txt

$ more out8.txt/part-00000 
(www.j1.com,126)
(tieba.baidu.com,65)
(113.105.174.5,60)
(www.baidu.com,54)
(mp.weixin.qq.com,46)
(119.147.106.188,40)
(bbs.caoav.net,31)
(i.ifeng.com,19)
(www.amazon.de,18)
(m.zhiyoula.com,18)
(www.360doc.com,16)
(br.pps.tv.iqiyi.com,14)
(www.1905.com,14)
(xa.btfs.ftn.qq.com,14)

下面对同一日志文件计算UV(Unique View)。

UV一般认为不同cookie的访问则算不同的独立用户。

package org.asiainfo

import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
import org.apache.spark.SparkContext._

/**
 * @author:zhaohf@asiainfo.com
 * @date:2015年1月27日 下午5:54:39
 * @Description: TODO
 */
object UniqueViewCount {

  def main(args: Array[String]): Unit = {
    if(args.length < 4){
      System.err.println("Usage:<input_file> <url_column_index> <output_file>")
      System.exit(1)
    }
    val conf = new SparkConf().setAppName("UniqueViewApp")
    val sc = new SparkContext(conf)
    val url_index = args(1).toInt
    val cookie_index = args(2).toInt
    val textRDD = sc.textFile(args(0))
    	.map(_.split("\t"))
    	.map(line => (new java.net.URL(line(url_index)).getHost) + "\t" + line(cookie_index))
    	.distinct()
    	.map(line => (line.split("\t")(0),1))
    	.reduceByKey(_ + _)
    	.map(item => item.swap)
    	.sortByKey(false)
    	.map(item => item.swap)
    textRDD.saveAsTextFile(args(3))
  }

}

spark-submit --class main.UniqueViewCount target/scala-2.11/spark_2.11-1.0.jar hdfs://localhost:9000/user/root/intput/1000_log 2 4 hdfs://localhost:9000/user/root/output

more result.txt
(bbs.caoav.net,31)
(www.baidu.com,28)
(www.amazon.de,15)
(m.zhiyoula.com,15)
(www.360doc.com,14)
(m.sohu.com,11)
(mp.weixin.qq.com,11)
(www.kaixin001.com,10)
(www.zhiyoula.com,7)
(lolbox.duowan.com,7)

下面用shell来验证正确性：

先用python解析出url中的host:

#!/usr/bin/python
from urlparse import urlparse
import sys
with open(sys.argv[1],'r') as f:
        for line in f.readlines():
                splits = line.split('\t')
                url,cookie = urlparse(splits[2]).netloc,splits[4]
                print url + '\t' + cookie

 $ python check.py 1000_log > 1000_log_pre

 $ cat 1000_log_pre | sort | uniq | awk -F '\t' '{print $1}' | sort | uniq -c | sort -nr -k1|  head
     31 bbs.caoav.net
     28 www.baidu.com
     15 www.amazon.de
     15 m.zhiyoula.com
     14 www.360doc.com
     11 m.sohu.com
     11 mp.weixin.qq.com
     10 www.kaixin001.com
      7 www.zhiyoula.com

使用Spark计算PV、UV

标签：spark

原文地址：http://blog.csdn.net/laozhaokun/article/details/43196425